turing-surface-cache

Reaction-diffusion (Turing patterns) on curved surfaces, evaluated at a chosen end time. Every setting is a choice from a short list, so each combination names one exact solution. Solutions already in the shared cloud cache load by themselves as you browse the choices; a combination no one has computed shows empty surfaces until you press Compute solution, which runs it here (with WebGPU, via turing-surface's spectral solver). Drag to rotate.

model

geometry

seed end time

Solutions computed here can be contributed back to the shared cache, so the next visitor gets them instantly. Contributing requires an upload API key.

upload API key

or on a machine with no browser on it:

if that command does not run on the other machine

Unexpected token {, or another syntax error, means that machine's node is too old to read the program at all: node 18 or newer is needed. nodejs.org has current builds, and nvm, fnm and asdf install one per user without touching what the system depends on; where a cluster uses environment modules, module load node may be all it takes.

the webgpu package is installed but did not load … GLIBC_2.34 not found means the machine's distribution is older than the prebuilt Dawn binary — ordinary on a long-lived Linux workstation, since Rocky and RHEL 8 have glibc 2.28 and Dawn wants 2.34. Running the command inside a container is the obvious answer and the wrong one: there the NVIDIA driver declines to bring up its Vulkan driver, while on the host it comes up fine. What does work is borrowing a newer userland from a container image and running node through that loader, still on the host, so the GPU, /dev and /proc are exactly what they were:

# 1. borrow a userland — podman or docker unpacks it; nothing runs inside it
mkdir -p ~/turing-userland
podman create --name u docker.io/library/node:22-bookworm-slim true
podman export u | tar -x -C ~/turing-userland
podman rm u

# 2. a node that uses it, running on the host rather than in a container
U=~/turing-userland; mkdir -p ~/turing-bin
printf '#!/bin/sh\nexec %s --library-path %s %s "$@"\n' \
  $U/lib/x86_64-linux-gnu/ld-linux-x86-64.so.2 \
  $U/usr/lib/x86_64-linux-gnu:$U/lib/x86_64-linux-gnu:/usr/lib64 \
  $U/usr/local/bin/node > ~/turing-bin/node
chmod +x ~/turing-bin/node

# 3. run the command above, with that node found first
export PATH=~/turing-bin:$PATH

The host's own /usr/lib64 stays last on that library path, and is where the NVIDIA libraries and the Vulkan loader come from: those have to match the running kernel module, so they must be the machine's own. If podman's storage lives on NFS it will fail to unpack the image, which --root /tmp/$USER/store --runroot /tmp/$USER/run on each podman command works around.

WebGPU is not available, or no adapter at all, means there is no GPU to reach: Dawn goes through Vulkan on Linux and Metal on macOS, so a machine without a GPU, or with one whose driver is not installed, has nothing to offer. DAWN_FLAGS='backend=vulkan' makes it say more about why.

A rate of a few steps per second rather than a few hundred means a software rasterizer has stood in for the GPU. The command names its adapter on the first line and says so when it notices, but it does not refuse to run — a slow contribution is still a contribution, if that is what you meant.